Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I have a series of N Bernoulli tests (p, 1-p).

I need to calculate a probability of passing more than N/2 tests, depending on N and p.

The obvious solution is Chernoff bound: $\varepsilon \leq 2^{-N(p-\frac{1}{2})^2}$, but this is not sufficient for me. I actually need some stronger dependency on p. Is there any available?

I tried fitting Hoeffding, Azuma and Bernstein's inequalities, but it looks like all of these also do not give any sufficient dependency on p.

Is there any convenient estimation?

What I need is something like: $\varepsilon \leq 2^{-N*p}$

share|cite|improve this question
Are you looking for a better expansion in $p$ for $p$ close to $1/2$ or for $p$ close to $0$? – Kevin Costello Oct 17 '12 at 17:48
p closer to 0. Like, passing a test is highly improbable. – user10732 Oct 17 '12 at 21:43

Did you try the relative-entropy version of Chernoff's bound?

Given $n$ samples, define $\hat p$ to be the empirical probability (i.e. successes divided by total trials). Then for any $q \in [0,1]$ we have

$$ P(\hat p \geq q) < \exp(-n * RE(q, p)) $$ where the relative entropy $RE(q,p)$ is defined $$ RE(q,p) = q \ln \frac{q}{p} + (1-q) \ln \frac{1-q}{1-p} $$

The same bound holds for $P(\hat p \leq q)$. In your case you would be interested in $q = 1/2$.

share|cite|improve this answer

If $p$ is tending to $0$, one option is just to use the union bound. There's $\binom{N}{N/2} \leq 2^N$ sets of size $N/2$, each of which occurs with probability at most $p^{N/2}$. It follows that the probability some set of $N/2$ events occur is at most $2^N p^{N/2}.$

This is in general not very tight (if you need any sort of a tight bound, working through a Chernoff bound like David Harris suggested is probably the way to go), but for some applications where you only need the probability to go to $0$ a crude bound like this works nicely.

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.